September: The State of Play (Defence)

This post is a follow up to one written earlier in the week which looked at Fulham’s attacking statistics: you can read that here if you missed it.

The focus of this piece is the defensive stats related to the first 3 gameweeks of the Premier League, ie up to and including Fulham’s defeat against Aston Villa.

As noted before, I am building a dashboard of these key stats and am intending to monitor how they evolve over the season, at the moment, with a sample size of 3 games, they will contain all sorts of random variation which will smooth out of the stats overtime. But at this moment, it gives the best overview that I can provide of the story of Fulham’s defence so far.

As before, I will lead with a summary of the observations, with more detail below for those who want to get more into underlying numbers.

Summary

Fulham have continued to deploy a very aggressive defensive system based on a high press, modelled on the one used to beat Brentford at Wembley…

they rank behind only Liverpool and Tottenham for the number of occasions they press opponents in the attacking third of the field
they are 5th in the league for successful tackles in the attacking third too

and this approach may be contributing to Fulham’s success in keeping down the volume of shots created by opponents…

Fulham have conceded just 34 shots, the same as they have created themselves and a close match to the league average

however, nearly a third of these shots conceded have found the back of the net…

we see that Fulham are giving up very high quality chances, and that these high quality chances are turning into even higher quality shots but even this does not explain the very high proportion of shots which are turning into conceded goals.
We observe that, historically, unsuccessful premier league defences (particularly those which end up relegated) tend to have the characteristic of conceding better shots than xG models predict. We can speculate that this is a function of the extra space that they afford shooting attackers.

when we look at how the goals are conceded, we see that, at this stage of the season, the majority are from either poor marking at set pieces, or basic individual errors in defence…

only 3 goals conceded were a result of the press being broken, by contrast:
a further 3 goals were due to a lack of marking on set pieces and
another 4 came from individual errors (not tracking players or mis-kicking the ball)

and this raises the possibility that, by improving the performance of defensive personnel (particular via transfers), Fulham could dramatically reduce the number of chances conceded to the point where they are competitive in matches.

Goals, Shots and xG

The most important measure of Fulham’s defensive effectiveness is the number of goals they concede, and in this respect, as we all know, Fulham are doing very badly. They are currently conceding 3.3 goals per game, with only West Brom doing worse (and surprisingly, just behind the two Manchester clubs).

So clearly we need to see if the stats say anything about why Fulham have conceded so many goals in these opening games.

The next key number I want to look at is the number of shots Fulham have faced, and this is interesting because it is not a particularly high number: 34 shots in fact (or 11.3 shots per game), which as you will see in the chart below, is pretty average and, for what its worth, matches exactly the number of shots Fulham have taken themselves!

So this suggests that Fulham have either been unlucky, or the shots they are giving up are more dangerous than those given up by other teams.

Unfortunately, the data suggests the latter explanation to be the best fit we can measure the quality of chances conceded using models of expected goals per shot.

In this measure, we see Fulham ranking lower in the league, at 15th with 1.7 expected goals conceded per game. Not a good number, but Newcastle, Man City, Man Utd, Leeds and West Brom all producing worse at this point in the season!

The implication of this higher level of expected goals conceded (herein referred to as xGA) is that the average quality of chances given up by Fulham is very high, and indeed it is. Fulham’s xGA per shot is 0.15, the 4th highest in the league, and it is a figure which is about 3 times higher than their xG created per shot (which as discussed in the attacking summary is the lowest in the league).

So Fulham are giving up an average number of chances, but the chances are relatively high quality.

And it gets worse, you may be noticing, at this point, that Fulham’s expected goals conceded of 1.7 is around half their actual conceded goals of 3.3!

So why is that? Why are Fulham conceding so many more goals than xG models predict, is it bad luck?

It is possible that this is the case, they could be the victim of ruthless but fortunate levels of finishing by opponents, the sample size is certainly too small to rule that out, but I suspect there is more to it that that.

Exploring this issue further, I have looked at Post Shot Expected Goals against (PSxGA), this is a measure that looks at the likelihood of a shot being a goal based on the trajectory of the shot itself, rather than simply the position and circumstances under which the shot was taken (which is what the standard xG model does).

On this measure, based on the quality of shots taken, Fulham should have conceded 2.5 goals per game, which ranks 19th in the league (matching their rank for actual goals conceded) but still gives a number around 0.8 goals per game less than they are actually conceding.

There are two important implications here:

Opponents are getting away significantly better shots than we would normally expect from the positions and situations they are shooting from, indeed the post-shot xgA numbers are around 50% higher than the normal xG numbers. This could be luck or it could point to something more worrying (which I will come back to shortly)
Fulham are still letting in more goals than they should given the quality of shots being taken, indeed, in total, they have conceded around 2.5 more goals than the model suggests they should have done. This could be due to bad luck, or it could reflect poor goalkeeping (and I pause to note Fulham’s collective GK save % is a rather meagre 44%). I would suggest it is far too early to speculate on which it is, but it is something I will come back to as the season develops.

Going back to the first issue (opponents are taking better shots than they should be given the chances they are creating), I am concerned by this as this is a characteristic of highly unsuccessful premier league defences. Indeed over the last 3 seasons, 8 of the 9 relegated teams demonstrated the characteristic of giving up better shots (measured by post PSxGA) than would be expected from the chances that created them (measure using standard xGA).

Fulham, for example, in 2018/19, had a PSxGA which was 15% higher than their xGA. This is the highest discrepancy of the 9 relegated teams I looked at and means that opponents are consistently taking better shots than xG models would predict. After 3 games in this season, this figure stands at a whopping 47% and this is highly alarming. It means teams are consistently taking much better shots than we would expect them to, given the circumstances under which they are shooting.

You may wonder why a team would have this characteristic of giving up better shots than predicted by xG models? Well one possible explanation is that they are simply worse at defending than the standard assumption that goes into a typical xG model (some or all of the discrepancy could be luck too).

To unpack that a bit more, it means that an xG model will assume that a premier league attacker, when shooting, is under a certain degree of pressure. This pressure constrains their choice as to which foot to shoot with, where on the goal they can aim the ball, their time to pick a target, their ability to study the goal and goalkeeper position, etc.

I would argue that these weak and relegated teams probably afford opposing attackers more time than stronger defensive teams and this results in their opponents consistently taking better shots than modelled simply because of a lack of effective pressure.

To support this, I would consider some of the goals Fulham conceded this season, I will give an example, although I think it applies to many others.

Patrick Bamford scored with his weak foot for Leeds’ third goal in the match in gameweek two. A (slightly blurry) image of the shot is shown below.

Infogol’s xG model rated this chance as a 6% probability of being a goal (and this may be in part because Bamford had a bad record on his right foot).

But in reality, all he needed to do (and did do) was roll this shot to the far post. Indeed the final shot goes about 4 feet inside the post, and still comfortably beats the keeper. I would argue that the fact that Bamford has an unopposed run into the box before this, and is under minimal pressure while shooting turns this into a much simpler chance than the model predicts. In other words xG models simply assume that when shots happen in the premier league, players are under more pressure than this.

There are lots of other examples, including Costa’s first goal from the corner in the same game. It was rated as a 7% goal probability, but I suspect this also reflects that xG models don’t expect a player to receive the ball from a corner, take a touch to control, set themselves to shoot and then get a chance to volley at goal, all on the edge of the 6 yard box! But Fulham let that happen!

We have also seen the opposite phenomenon impact Fulham as well. In my article on Fulham using 5 at the back to see out games, we discussed that Fulham seem to massively outperform their xGA when using the 5 at the back system. Many readers suggested (very plausibly) that the extra bodies in defensive positions meant that attackers had reduced time and space and that the xG models were over-rating the chances conceded. If that is the case, then it is plausible that now Fulham’s lack of robust defence is causing the same models to under-rate the chances, and in both cases this would explain the data anomalies that we observe!

Power of the Press

I will now focus on Fulham’s tactical approach to defence. In the first two games of the season, Fulham adopted much the same model as used against Brentford at Wembley, that is a 4-2-3-1 formation.

In this system the front 4 are instructed to press the ball, often quite high up the pitch. Their objective is to either to win the ball directly or to force a pressurised and inaccurate pass downfield to be mopped up by the mid-field.

The diagram below gives an example of this system in effect, with a front 4 applying pressure to the ball high up the field. The obvious weakness in this approach, as I have described it, is that the central midfielders have too much space to cover, and it would be relatively easy for the team in possession to break the press by passing through to the mid-field, particularly in the wide areas highlighted in the diagram below.

The solution to this problem, which Fulham have used, is to push the fullbacks into the ‘mid-field’ area, to fill the highlighted gaps. This is shown below:

This prevents opponents easily escaping the 4 man press, but as shown above, it creates a new area of vulnerability in behind the advanced fullbacks and leaves the centre-backs fairly exposed. An example in the diagram below shows how the weaknesses of this system can be exposed, by playing attacking wingers into the gap behind the fullbacks.

In order for the above system to work then, it is critical that when the ball beats the pressing front 4, the next layer of players do not give the attackers room to turn and pass forwards, as if they do, they will be able to thread passes into the spaces shown above. Harrison Reed is very good at this and you will see him working hard to prevent opponents getting the space to pick passes of the type shown.

Nonetheless, we have conceded goals on a number of occasions this season, when opponents have managed to attack these gaps in the full back areas. I believe Parker tried to combat this risk against Aston Villa by sacrificing one of the pressing group (turning it into a pressing 3) in order to have a back three which was more able to defend the highlighted zones (I don’t think, as some have suggested, that he figured the best way to strengthen central defence was to put on more centre backs).

The above sets out the defensive gameplan for Fulham. We can now look at some of the stats to get an idea of if it is working!

Defending by Numbers

The table above summarises some of the defensive measures available, I have shown the raw metric (on a per game) basis and then shown how that compares to other teams in the league.

Most of the numbers look fairly middling, tackles attempted and won, for example, look encouragingly mediocre (ie they are not awful and are about average for the league).

The numbers that stand out to me are those that tell us something about pressing, Fulham rank 3rd for the number of times they apply pressure in the attacking third of the pitch, only Liverpool and Spurs do it more.

They are also 5th in the league for tackles won in the attacking third, and these numbers suggest that to some extent, the pressing strategy is effective. We also note that when Fulham apply pressure to the opposition, they rank 13th (which I would say is OK) for the percentage of times this leads to recovering possession.

To pull together the above into a consistent narrative I would offer the following:

Fulham are applying a very aggressive pressing system, more so than the majority of other teams.
This approach is doing a reasonable job in winning possession for Fulham and limiting the number of chances opposing teams have.
However, there is a risk that the system exposes the centre-backs by creating gaps behind the full-backs for opponents to exploit.
We also can see that Fulham appear to be giving up really high quality chances, and that these chances are leading to even higher quality shots on goal!

So the question I am wondering is whether the pressing system, and the space it creates to expose the centre-backs when it gets broken, is the reason why the relatively small volume of chances given up are proving so deadly to Fulham?

I think it is probably too early to say, but we can keep an eye on this issue as the sample size of games increases.

At this point, of the 10 goals conceded to date, I think 3, the third against Arsenal, and the third and fourth against Leeds are clear examples of a broken press leading to exposure of the space behind the full-backs.

A further 3 (second against Arsenal, first against Leeds and third against Villa) were botched set-pieces, ie a failure to mark properly. I dont think those were ‘system’ or tactical problems.

The remaining 4 I think fall into the category of individual error. Ream’s mis-kick against Arsenal, Bryan’s push on Bamford and Odoi and then Tete’s failure to pick up the run of Grealish against Villa.

Final Thoughts

The problem at the moment is that Fulham are giving up a relatively small number of chances but a really strangely high proportion of these are turning into goals.

This could be bad-luck, or it could be that Fulham’s defending is so bad, it is breaking the xG model! Or some combination of both.

Nonetheless, the underlying numbers describing Fulham’s defence, i.e. shot given up, xG given up, press effectiveness, are not too bad and it is possible that by eliminating/reducing defensive errors and improving set piece defence, Fulham could get the goals conceded figure to match more closely to the shots given up and also reduce the quality of chances given up.

It feels at the moment that Fulham’s defensive results are so much worse than the underlying numbers suggest, that some improvement (reversion to the mean) must occur, but I also think that strengthening personnel in the transfer window will be critical. Not least because Scott Parkers defensive strategy does require centre-backs who can cope in short handed situations, and this is something which, at the moment, does not seem to be something they look capable of.